Descriptions:
Abridge, the clinical AI company whose software now listens to more than 100 million doctor visits, is the subject of this in-depth crossover episode between the Latent Space and Unsupervised Learning podcasts. Janie Lee and Chai Asawa — both senior leaders at Abridge — explain how the company has evolved from an ambient documentation tool into what they call a clinical intelligence layer, designed to operate before, during, and after patient-clinician conversations.
The discussion goes deep on the practical challenges of deploying AI in high-stakes healthcare environments. The guests describe a three-level personalization hierarchy: individual clinician preferences (down to whitespace formatting), specialty-level calibration (a cardiologist’s note looks nothing like a dermatologist’s), and health-system-level integration of proprietary clinical guidelines. They explain how Abridge’s CEO, a practicing cardiologist who still rounds once a month, provides rapid internal feedback, and they discuss the difficulty of building offline and online evaluation pipelines for specialty-specific output quality. A key theme is the concept of AI-generated “slop” in medical notes — output that lacks the patient-specific context that makes documentation clinically meaningful — and how Abridge’s flywheel of edit data helps the system improve over time.
Asawa also previews the company’s expansion into clinical decision support, describing a vision where accumulated patient context, practice guidelines, and medical literature combine to proactively surface insights to clinicians at the moment they matter most — what she frames as moving from reactive alerting to proactive intelligence.
📺 Source: Latent Space · Published May 14, 2026
🏷️ Format: Interview
